A security-focused skill for sanitizing external content and neutralizing prompt injection attacks in AI agent workflows.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install guardian-wall-azzar
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install guardian-wall-azzar using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Guardian Wall?
Guardian Wall serves as a critical security layer for developers using Openclaw Skills. It specializes in identifying and mitigating both direct and indirect prompt injection (IPI) attacks, which often hide within external web content, user-uploaded files, or third-party data streams. By integrating this skill, users can ensure their AI agents remain within defined operational boundaries even when processing untrusted text.
This skill is essential for maintaining the integrity of Openclaw Skills, providing automated sanitization routines and auditing protocols that detect malicious instructions such as system overrides or hidden persona changes. It transforms raw, potentially dangerous input into isolated, safe data blocks through a combination of pattern matching and sub-agent verification.
Guardian Wall Use Cases
- Processing untrusted text fetched from external URLs or web scraping to prevent system manipulation.
- Analyzing user-uploaded documents and files for hidden malicious instructions like ignore all previous instructions.
- Protecting high-stakes workflows where prompt integrity is mission-critical for the application.
- Auditing suspicious content using dedicated sub-agent verification to detect hidden intent.
- Implementing standardized defense layers across various Openclaw Skills to maintain a unified security posture.
How Guardian Wall Works
- Input Sanitization: The scripts/sanitize.py utility is executed to strip non-printable characters and zero-width spaces that frequently hide injection vectors.
- Pattern Detection: The system scans the input for high-risk strings like Administrative access or System override based on a comprehensive reference database.
- Sub-agent Auditing: For complex or high-risk inputs, a specialized sub-agent is spawned to evaluate the text for hidden intent to manipulate the agent.
- Block Isolation: Sanitized content is wrapped in unique, randomized delimiters using specific hash-based XML tags to prevent the LLM from confusing data with instructions.
Guardian Wall Setup
To integrate Guardian Wall into your Openclaw Skills environment, ensure you have the necessary script permissions and reference files configured:
# Navigate to your skills directory
cd guardian-wall
# Ensure the sanitization script is executable
chmod +x scripts/sanitize.py
# Run a test sanitization on a sample file to verify the installation
python3 scripts/sanitize.py --input untrusted_content.txt
Guardian Wall Data Schema & Taxonomy
| Component | Description | Format |
|---|---|---|
| scripts/sanitize.py | Main sanitization logic and pattern detector for untrusted text | Python Script |
| references/patterns.md | Comprehensive database of known injection vectors and bypasses | Markdown Reference |
| Isolation Delimiters | Randomized hash-based tags used to wrap external content (e.g., EXTERNAL_BLOCK_[HASH]) | String Tag |
Guardian Wall Advanced Features
- Sub-agent Auditing: Automatically triggers a secondary AI model to perform a forensic analysis of suspicious text to detect manipulation.
- Randomized Sandbox Delimiters: Uses dynamic hash-based tags to prevent attackers from predicting and escaping isolation boundaries.
- Hidden Text Detection: Identifies CSS-based injection attempts such as display:none or zero-size fonts commonly used in web-based attacks.
- Extensive Pattern Library: Includes a frequently updated repository of known prompt injection bypass techniques specifically for Openclaw Skills.
SKILL.md
Loading
Related Openclaw Skills
File Organizer Skill
Automatically sort cluttered directories into organized subfolders by file extension or creation date.
clinstagram
A comprehensive hybrid CLI for Instagram that bridges Meta Graph API safety with private API feature depth for AI agents.
Daily Report Generator
A tool that automates the creation of professional work reports by aggregating data from git commits, calendar events, and task managers.
OpenClaw Safe Change Flow
A robust workflow and automation script for making safe configuration changes to OpenClaw instances with automatic rollback capabilities.
Human Writing
A specialized skill designed to eliminate AI stylistic tells and produce high-density, professional technical documentation.
Mema Brain
A standardized memory system that combines a SQLite metadata index for document organization with a Redis-powered short-term context buffer.
Featured*
